perm filename NSF.E[DOC,TOB] blob
sn#027211 filedate 1973-02-26 generic text, type T, neo UTF8
00100 ABSTRACT
00200
00300
00400 The Stanford Artificial Intelligence Laboratory proposes to undertake
00500 a program in advanced automation over
00600 a period of one year at a cost of $143K. This proposal is to
00700 continue and extend programs sponsored under ARPA Contract SD-183.
00800
00850 This proposal is for initial study and preliminary research on a
00860 three year "proof of concept" project in computer controlled automatic
00870 assembly. The outline of the project has been worked out
00880 with industrial and governmental advice, but more analysis as well
00890 as more research is needed.
00895
00900 The project includes a pilot assembly station carrying out practical
01000 assembly tasks, a hardware-software system ready for production, and
01100 fundamental work on the underlying problems of vision, manipulation
01200 and problem-solving. We believe the laboratory is uniquely qualified
01300 to carry out this program which will be an important contribution to
01400 increasing industrial productivity.
01500
01600 Our basic approach involves a general purpose system integrating
01700 sensors and manipulators controlled by sophisticated programs in a
01800 digital computer. We have worked with such systems since 1965 and
01900 have developed many of the basic ideas required in general purpose
02000 assembly devices. The main thrust of this proposal is to establish
02100 the practicality of our methods and carry their development to a
02200 point from which industry can take over.
02300
02400 The research program is divided into four main sections: Pilot
02500 assembly station, vision development, manipulation development, and
02600 systems support. Each section contains a brief description of the
02700 laboratory's past work in the area, how it is to be applied, and what
02800 extensions are to be undertaken. Finally, we consider a number of
02900 important issues concerning our relations with industry and with
03000 other reseach and development efforts in advanced automation.
03100
03200
03300 TABLE OF CONTENTS
03400
03500 Part Page
03600
03700 I The Need for Advanced Automation ...................... 1
03800 II Research Plan ......................................... 4
03900 A. The Pilot Assembly Station .................... 4
04000 B. Research in Vision ............................ 6
04100 C. Manipulator Development ....................... 9
04200 D. Systems Requirements for Advanced Automation .. 10
04300 III Dissemination of Results and Relation with Industry ... 12
04400 IV Staff ................................................. 13
04500 V Computing Facilities .................................. 22
04600 VI Budget ................................................ 23
04700 VII References ............................................ 26
04800 VIII Relevant Laboratory Publications ...................... 30
04900 IX Films ................................................. 32
05100 X Cognizant Personnel ................................... 49
00100 I. THE NEED FOR ADVANCED AUTOMATION
00200
00300
00400 The contributions that advanced automation can make to enhance productivity
00500 in manufacturing are now widely
00600 appreciated. Current interest in advanced computer-oriented
00700 automation stems from two sources. The first is the decreasing
00800 cost of computation combined with an improved understanding of how to
00900 apply computation to various aspects of manufacturing. The second
01000 cause is the realization that many current manufacturing operations,
01100 primarily assembly and inspection, will require complex
01200 automation techniques which have are now research techniques. The same considerations apply to the
01300 automation of tasks such as laboratory and medical procedures, but
01400 the emphasis has been (as ours will be) on manufacturing.
01500 A recent ARPA-sponsored study
01600 [Anderson 1972] in which we participated suggests the following major
01700 benefits from advanced automation:
01800 improved national productivity
01900 reduced prices
02000 increased product quality and uniformity
02100 customization of products
02200 increased job satisfaction.
02300 We would like to add just one comment. As we have seen with other
02400 technological developments such as numerical-controlled tools, there
02500 will be things made which were previously impractical to make. There
02600 is widespread agreement in the technical, industrial and government
02700 communities on the potential benefits of advanced automation. There
02800 is much less agreement on the level of sophistication needed in
02900 automation hardware and software. While there are large application
03000 areas in which specially constructed machines or simple general
03100 purpose devices will prove optimal, we believe that the most
03200 practical solution for intermediate quantities (500 to 100,000) will be to develop
03300 general purpose assembly and inspection devices controlled by
03400 sophisticated programs in digital computers.
03500
03600 Several studies have indicated some priorities among directions of
03700 automation. The ISI study [Anderson] indicates that less than 20%
03800 of discrete manufacturing is amenable to special purpose automation.
03900 A Chrysler Corporation study indicates that car assembly labor
04000 has first priority in automation; assembly plant direct labor is
04100 40% of direct corporate labor. It was felt that automation was a
04200 must in the small parts division. The ISI study quotes a figure of
04300 75% as the fraction of batch produced discrete manufactures. One
04400 conclusion is that "the emphasis in development should be on
04500 simple vision devices and on low-bandwidth tactile and force
04600 sensing feedback". That study indicates that the assembly process
04700 typically accounts for 50 percent of production cost. Although
04800 such estimates are crude, the study indicates a potential savings of
04900 $27 billion of manufactures which total $492 billion.
05000
05100 We assume that present manipulators have a small but stable market,
05200 and that most will have minicomputers soon (Sundstrand does now).
05300 Will the incremental cost of touch, force and visual sensors be
05400 offset by the opening of new capabilities, and will the cost be
05500 offset by reduced cost of system design for machines which allow
05600 loose alignment tolerances?
05700 We estimate that touch and force sensing will add about 5% to the cost
05800 of current manipulators of the Unimate class (assuming a multiple of
05900 two between cost and selling price).
06000 We estimate that by 1977, the
06100 incremental cost of visual sensors, a laser ranging device such as we
06200 use now,and additional memory requirements, with the multiple of two in
06300 marketing, will add about 15%-20% to current manipulator prices. If
06400 labor costs continue to rise, part of the increase will be offset,
06500 but cost comparisons will not change appreciably over that time scale
06600 unless significant new mechanical designs are made. A conclusion
06700 is that sensor-equipped robots will be available at slight additional
06800 cost. We explore the benefits of incremental improvements to
06900 manipulators.
07000
07100 We do not expect mechanical production costs to drop as dramatically
07200 as electronics costs fall with integration. The most promising cost
07300 reduction development would be high-production standard modules such
07400 as robots. As an example, consider the Stanford arm; produced in
07500 unit quantity, parts, manufacturing and assembly costs are
07600 approximately $10K. The arm is much simpler than an automobile;
07700 there is no reason why it should cost more. A smaller, more recent
07800 design sells for $3500 in quantities of two.
07900
08000 Potentially, a robot with two arms, minicomputer control, force and touch
08100 sensors can be constructed for under $10K in quantities of 1000
08200 now.
08300 Details are shown later. To summarize: two arms
08400 (new small Scheinman design) $2000 with hand; arm drivers and arm interface
08500 electronics $1000;
08600 force and touch $800; small computer 8k memory $4500; total $8300.
08700 Assuming that the computer price falls to $1000 by 1978, the cost is $5000.
08800 Of course, these are small arms, suited for parts assembly, not automotive
08900 assembly.
09000 This is the
09100 range of price of present transfer machines which have limited market
09200 because they have limited motion, no control, and are thus locked out
09300 of the economy of scale. There is a dynamic which leads to economy of
09400 scale; the greater the versatility, the wider is the market for
09500 applications. This will be true particularly in small production and
09600 job shop operations which have been difficult to automate. The
09700 larger the production, the more versatility for the same initial cost
09800 as special devices. Also, with high volume, standardization of
09900 maintenance, and availability of parts and service are improved. The
10000 quality of the device is also improved.
10100
10200 The economies of scale in production of assembly modules are matched
10300 by economies of intellectual scale in development of design for
10400 automation. The design process is modularized and easily
10500 communicable; there is a common discipline which can be widely
10600 absorbed; developments of many sources can be integrated.
10700 Programming and training done once for one machine can be transferred
10800 to all machines in a plant, and even transferred nationwide.
10900 Functional modularity has been emphasized in our studies. Modularity
11000 is clearly possible and there is a powerful economic incentive to
11100 assemble widely varying systems from one, two, or three arm modules,
11200 force and touch modules, vision and ranging modules, with planning
11300 and monitoring modules. This would also entail software modularity
11400 with computer-aided mechanical design and job specification for
11500 robots on a local or network time-sharing service.
11600
11700 The alternative is special design for each operation. We expect that
11800 at some point, design will be cheaper with standard robot modules
11900 than with special purpose designs, just as it is now more economical
12000 to use mini-computer controllers than special purpose controllers.
12100 Modularity has the potential to reduce design time and improve
12200 predictability of design time and costs, permitting better planning
12300 and response to competition. To some extent, the dehumanized
12400 industrial worker now fills the role of standard module, so that we
12500 do not expect discontinuous change. The continuity with decades of
12600 experience with design for human versatility in assembly can be
12700 relied upon; robots with some versatility, sensing, contingency
12800 actions, and dexterity will provide a transition to newer schemes
12900 which will evolve more slowly. Another advantage of general purpose
13000 techniques is the ease with which the size of the robot can be scaled
13100 up or down. There appears to be no difficulty in making very large or
13200 very small assembly stations which use essentially the same programs.
13300
13400 Versatile assembly devices have an efficiency advantage in medium
13500 scale and job shop production; their utilization efficiency can be
13600 higher because versatility allows a variety of operations to be
13700 scheduled at one station. The availability of computers for
13800 production scheduling will allow the flexibility to be used to
13900 advantage. Robots would also have longer life over which to amortize
14000 costs, since they could function with greatly changed production
14100 operations, and evolve with new hardware and software modules.
14200 They would have higher reliability, since they would have monitoring,
14300 contingency action, and error recovery. They would be self-aligning,
14400 self-calibrating, and self-testing.
14500
14600 With simple devices, transfer and assembly operations must be broken
14700 down into small operations with transfer operations between.
14800 This introduces extra transfer devices with orientation-preserving or
14900 re-orienting steps, and overlapping operations which can be combined
15000 in an advanced device. The extra cost of jigs and fixtures, of stopping
15100 assembly lines adds to the cost of special purpose automation.
15200 Robots can cut down on real estate, since
15300 several operations can be done at one station.
15400
15500 Because of the adaptibility of work stations, a quicker response to
15600 market and competition is possible. Production runs can be smaller,
15700 with less capital tied up in inventory and unused machines. There
15800 can be better response to the needs of high technology industry which
15900 often requires small quantities of instruments. There has been
16000 difficulty in meeting the tight tolerances necessary for operation of
16100 non-intelligent devices. It is hard to estimate the value of
16200 systems which work within current industrial tolerance, rather than
16300 requiring radical improvements in engineering practice, but that
16400 continuity does seem important.
16500
16600 Serious issues arise in the development of robots with simple sensory
16700 systems. Some evaluation issues should be settled by demonstration,
16800 since much sharper analyses can be made this way. Demonstrate practical
16900 speed and dexterity with two arms and tool using. Demonstrate the
17000 generality of arm control and planning for various arms and devices.
17100 Demonstrate designs which standardize interface with devices and
17200 which lower interface costs. Demonstrate programming of assembly
17300 devices in an interactive APT-like system (develop the Stanford
17400 Assembly Robot Management System, ARMS); show the commonality of
17500 operations, ie solving a few tasks gives most of the modules to solve
17600 the next task.
17700
17800
17900 What is the current involvement? GM Research Lab has demonstrated
18000 a wheel mounting operation using visual alignment. IBM has a small
18100 robot project. Hitachi has demonstrated a robot with touch and a
18200 system with visual identification of objects on a conveyor belt,
18300 with visual control of manipulation. Can industry be expected to
18400 adequately support this development? Development of primitive
18500 robots is an evolutionary process, not a single product development.
18600 Thus it is not possible to keep secret and maintain an enormous
18700 productivity advantage; the techniques are usable by everyone. The
18800 area is too big; the payoffs are huge and the investment level is
18900 large. Also, traditionally, industry has been dependent on suppliers.
19000 But with the exception of IBM, the suppliers are smaller and less
19100 able to risk the development. Management techniques require inordinate
19200 return on capital.
19205 An alternative support could come again from the military. The ARPA
19210 study and the history of numerical controlled machine tools are natural
19215 consequences of the vast military procurement in technologically advanced
19220 products. In the case of N/C, funds could be diverted from an outmoded
19225 technology to development of devices which were essential to production
19230 plans for the future. It is high technology products which are most
19235 competitive internationally in the civilian economy. The record of
19240 government investment in numerical control, solid state electronics,
19245 computers, and electro-optical devices has been very productive.
19250 However, in large scale development of civilian products and international
19255 markets, foreign governments have taken the lead.
19300 The development of general purpose assembly devices will benefit
19400 manufacturers in a broad spectrum of industry, particularly medium
19500 scale manufacturing. It is an appropriate area for governmental
19600 action to help meet a national need.
19700
19800 Given sensor-equipped robots at little extra cost, they will have
19900 visual ability to sense color, strong contrasts in brightness,
20000 recognize known objects from two dimensional outlines (top view of
20100 flat object) or three dimensionally using depth, to a certain extent
20200 they will be able to recognize three dimensional objects using
20300 monocular recognition. They will be able to visually align mating
20400 parts in assembly operations, increase reliability by self-
20500 calibration and flaw and disaster detection, and permit large
20600 tolerances in parts transfer, even permitting parts to be taken from
20700 a bin. Several of these abilities are not yet available in the
20800 laboratory, but with known conditions and objects, and some
20900 engineering (e.g. color coding) are achievable within the time span
21000 of the proposed project.
21100
21200 The major determinant of the economic impact of assembly robots will
21300 be the extent of their capabilities. Assessing and extending these
21400 capabilities is the principle goal of the proposed project.
21500
00100 II. RESEARCH PLAN
00200
00300 A. THE PILOT ASSEMBLY STATION
00400
00500 The basic goal of the proposed project is to implement a general
00600 purpose assembly station and to prove its applicability to real
00700 problems. The project will be considered a success if one or more
00800 enterprises chooses to manufacture and market the general purpose
00900 station. There are a number of hardware and software problems to be
01000 solved in meeting our goal.
01100
01150 In one configuration,
01200 the assembly station is assumed to be connected to a large
01300 time-shared computer facility. Not only does this appear to be a good
01400 design for our problem, but it is also necessary for an integrated
01500 programmed automation facility. The only emphasized statement in the
01600 ARPA study report [Anderson 1972] is: "The maximum impact of
01700 computers in manufacturing will come from complete, real-time
01800 computer cognizance and control of all processes and resources,
01900 allowing the precise scheduling and allocation of these resources."
02000
02100 Each assembly station will consist of a small digital computer, one
02200 or more manipulators and cameras. Manipulators and cameras will be
02300 combined as modules in systems to suit a given task, and will be
02400 digitally controlled by the station computer.
02405 The station computer may be connected to a central computer or may stand
02410 alone. The latter configuration is less efficient in the long run,
02415 but allows for incremental automation.
02500
02600 The program in the station computer will be flexible as to the number
02700 and position of manipulators and cameras. It will control input from
02800 the cameras, provide for some pre-processing of the picture and the
02900 application of simple tests to the visual data. In the case of
03000 manipulators it will provide for simultaneous servo control. It will
03100 also process other inputs, such as from touch sensors on the
03200 manipulators, from photocell interrupt devices, limit switches, etc.
03300 Both analog and binary outputs will be available for the control of
03400 other devices.
03500
03600 Each station computer will have access to a task program specifying
03700 the actions and tests it should perform to carry out its task. This
03800 task program will be compiled by the central computer. Compilation
03900 will consist of the analysis of predicted scenes in order to specify
04000 simple tests that the station computer can carry out, the calculation
04100 of collision free trajectories for the manipulators, the sequence of
04200 operations and simple contingency plans. etc. We are planning for
04300 tasks whose normal execution does not require detailed scene
04400 analysis. If scene analysis becomes necessary (e.g. a dropped part)
04500 we envision the station sending an image to the central computer for
04600 processing.
04700
04800 The central computer will perform overall control of the individual
04900 assembly stations and the relative rates of production. If at any
05000 station a planned test fails, for which a contingency plan does not
05100 exist, or any unexpected sensor signals are received, the central
05200 computer will also be interrupted and appropriate actions taken.
05205 If the assembly station is not connected to a central computer, these
05210 operations will require human intervention.
05215
05300
05400 Most of the hardware required for the assembly station has been
05500 developed for the laboratory. Manipulators and cameras will be
05600 discussed in detail below. The major additional problem is to refine
05700 the existing designs to meet the demands of an industrial
05800 environment.
05900
06000 We are now acquiring small computer such as the PDP-11 and will
06100 interface the existing arms to it together with one or more cameras.
06200 A software package for the small computer will then be developed in
06300 order to control the manipulators, read the TV cameras and to perform
06400 low level visual analysis. Programs to plan tasks, calibrate the
06500 system and to perform higher level visual analysis, error recovery
06600 and strategy planning will be written in a high level language for
06700 the time-sharing processor.
06800
06900 The most difficult part of the assembly station, and the heart of the
07000 project is a demonstration of the feasibility of the station.
07100 We plan to do this by developing programs which actually assemble and
07200 inspect a variety of fairly complex manufactured items.
07300
07400 A typical example is our current effort, assembly of an automobile
07500 water pump. The major parts are to be located, the screws are
07600 supplied by a feeder, as is standard. Locating the body and cover of
07700 the pump require a gross localization, then a fine localization of
07800 screw holes. Touch provides a final accuracy and independent
07900 coordination of the hand. The placing of alignment pins within
08000 screw holes could be monitored visually; although in this case,the
08100 previous actions provide adequate accuracy. Placing of the cover
08200 over alignment pins can be carried out without vision. Improved
08300 force sensing now under development will be useful for that
08400 operation.
08500
08600 As we move to more difficult tasks, the sophistication required goes
08700 up quite rapidly. The current manipulation routines seem to be
08800 expandable to cover anticipated problems, especially since we intend
08900 to employ special purpose manipulators (tools) for some tasks. The
09000 main part of our effort will be vision for assembly; an assembly task
09100 will be performed using vision, touch, and two hands. The vision
09200 effort will be integrated with other abilities, and while no great
09300 generality will be stressed, new modules will be incorporated into
09400 the system, and a set of techniques will evolve. These will depend
09500 heavily on techniques similar to those of our recent studies in
09600 visual feedback and hand/eye coordination [Gill 1972]. The internal
09700 model of the world will be used to predict visual appearance of
09800 selected areas of the scene, and a set of verification procedures
09900 will confirm or contradict the predictions. Many of the operations
10000 will be connected with supplying feedback to manipulation, for
10100 alignment and insertion. These visual procedures will have models for
10200 the parts and approximate locations.
10300
10400 This and other visual tasks connected with assembly will initially be
10500 performed as special case problems. In addition, to make these
10600 techniques available, we will pursue development of strategy
10700 programs. This step is essential if intelligent assembly devices are
10800 to be conveniently programmed for complex tasks. Within the division
10900 of labor between large and small machines, this facility must reside
11000 in the central time-sharing facility which has large memory, file
11100 storage, and problem-solving languages. The system must have a model
11200 of the parts for assembly,and a sequence of instructions for
11300 assembly. It will be necessary to have a representation of shape and
11400 models for usual assembly operations such as insertion and alignment,
11500 and the ability to solve simple problems in connection with these
11600 operations, all this to enable convenient communication with the
11700 user. When these facilities exist, the addition of new features and
11800 operations can be incorporated smoothly, so that users can call on a
11900 growing library of planning modules. The result of a planning effort
12000 is a cycle of machine planning and man-machine interaction which
12100 results in a plan network for the mini computer which controls
12200 individual machines.
12300
12400 B. RESEARCH IN VISION
12500
12600 Our past work on visual feedback [Gill 1972] in stacking objects and
12700 inserting objects in holes is directly relevant to vision for
12800 assembly. The system provides feedback for picking up, then at the
12900 final position, with the hand still holding the block and in view,
13000 predicts the appearance of edges and corners within a small window
13100 around the faces to be aligned, and uses a corner/edge finder to
13200 determine alignment error. Incremental motion capability of the arm
13300 is used for error correction. That work incorporates
13400 self-calibration to coordinate hand and eye coordinate systems and
13500 provide for reliability, as well as to limit search by accurately
13600 predicting expected features. We have also found it useful to
13700 predict the appearance of complex objects and to input models of them
13800 at the console. Well-developed systems exist for both prediction and
13900 modelling [Baumgart 1972 a,b].
14000
14100 We have a set of visual operations which provide confirmation of
14200 predictions and direct interaction with the visual image. Several
14300 preliminary color region analysis programs have been developed
14400 [Binford, unpublished, Bajcsy 1972]. An edge operator [Hueckel 1969]
14500 and a region-based corner/edge finder [Gill 1972] both operate on
14600 small windows of the scene, as directed by a higher level. An edge
14700 verifier is used to confirm or reject edges predicted from context
14800 [Tenenbaum 1970]. Color identification has long been demonstrated.
14900 Recently, a program has been implemented based on Land's Retinex
15000 model of color constancy in vision, which defines a color triangle
15100 from ordering of brightnesses. The center of the triangle is
15200 designated as white. The program shows a color constancy similar to
15300 humans; table top scenes are assigned colors reasonably well under a
15400 variety of lighting with different spectra: sunlight, flourescent,
15500 and incandescent.
15600
15700 These techniques have been used on scenes more complex than
15800 polyhedra. Outdoor scenes were analyzed with a combination of color
15900 and texture region analysis [Bajcsy 1972]; depth cues were taken from
16000 the gradient of texture. Natural objects and natural scenes have been
16100 used for our work with motion parallax [Nevatia in draft] and stereo
16200 vision [Quam,Hannah].These are complementary techniques of depth
16300 perception which are applicable for planetary exploration and
16400 industrial tasks. Motion parallax can be used for planning of
16500 navigation and characterizing large scale terrain features, since it
16600 provides large baselines. Visual feedback with stereo vision is much
16700 more powerful than without; we have a program which has demonstrated
16800 stereo visual feedback.
16900
17000 We have been working with visual description and recognition of
17100 complex objects; we developed a laser ranging device [Binford 1970,Agin]
17200 which was operational two years ago. Using depth data and a
17300 representation of shape which we have developed, we have been able to
17400 describe the parts of objects like a doll, snake, and glove [Agin
17500 1972]. The representation [Binford 1971] is based on segmenting
17600 objects into parts connected at joints. Parts are described by a
17700 volume representation with arbitrary cross section defined along a
17800 space curve. Techniques have been developed which join parts into
17805 descriptions of entire objects.
17900
18000 A study of visual and depth ranging techniques has been carried out
18100 which provides a background for choosing sensors and interface, and
18200 making cost estimates [Binford 1972, Earnest 1967]. We have
18300 maintained a calibrated hand and vision system for several years
18400 [Sobel 1970], and have developed self-calibration techniques with one
18500 and two eyes [Yakimovsky unpublished, Hannah 1972, Gill 1972].
18600
18700 A system for understanding scenes of blocks from single views has
18800 been implemented [Falk 1970]; it segments objects, matches against
18900 internal models, predicts appearance from its assumptions, and
19000 searches for verification and contradiction based on the predicted
19100 image. Support and spatial relations are also obtained. There has
19200 been past work on programs to obtain line drawings as input to
19300 recognition systems. An earlier system was based on limited context
19400 of expecting continuation of edges at vertices and predicting missing
19500 edges in certain expected configurations [Grape 1970]. A more recent
19600 approach has been to use models within the process of generating line
19700 drawings, so that the most efficient next guess can be made and
19800 context used intrinsically in the formation of line drawings.
19900
20000 VISUAL DEVELOPMENT FOR ASSEMBLY
20100
20200 In the pump assembly, visual tasks consisted of gross location of
20300 large parts, then using the information to predict location of images
20400 of holes, and monitoring of insertion into holes. These tasks are
20500 typical of the simple visual operations which we will implement or
20600 adapt. We will greatly expand our efforts with visual feedback. At
20700 present, visual feedback techniques are limited to polyhedra. We will
20800 improve image prediction techniques to predict images of objects with
20900 curved edges. We will develop a simple region analysis program which
21000 assumes adequate contrast to locate the curved edges predicted within
21100 a small window. Color will be used when necessary. We will
21200 implement two-dimensional matching of boundary curves with predicted
21300 curves. These will be combined with our existing edge operator and
21400 corner/line operator. We will maintain a calibrated system and
21500 improve calibration updating so that reliability is maintained at a
21600 high level.
21700
21800 VISION RESEARCH
21900
22000 We will improve the laser hardware so that it can be used widely and
22100 extend our work with representation of complex objects using depth
22200 data. We are able to make complete descriptions of objects which
22300 consist of a single primitive segment, such as a torus, a snake or a
22400 cone. The programs can segment dolls, toy horses, in intuitively
22500 natural ways [Agin 1972]. It is necessary to improve our segmentation
22600 process in order to proceed to joining primitive segments at joints
22700 and make complete descriptions of complex objects. When we have
22800 these descriptions, the computer will have symbolic internal models
22900 of objects. No recognition has yet been done, although we are in a
23000 position to recognize those objects which have only one primitive
23100 segment, for which we can make complete descriptions. For recognition
23200 of more complex objects, once we have complete descriptions, we will
23300 encode the basic properties of the representation within a visual
23400 memory scheme which allows access of hypothesized similar
23500 three-dimensional shapes, with associated surface information, to
23600 compare more closely with extensive verification. We intend to apply
23700 large parts of our programs for description and recognition of curved
23800 objects using only monocular intensity information. This is a much
23900 harder problem, but within our representation, shares much of the
24000 same code, and relies on the same conceptual basis of local
24100 three-dimensional models.
24200
24300 We feel that progress in application of robotics depends on
24400 development and system use of representation. To make robots useful
24500 for batch production, programming must be rapid enough to change jobs
24600 rapidly. Ultimately, representation of parts and operations is
24700 crucial. Our representation primitives resemble some fabrication
24800 operations and provide a useful basis for description of shape.
24900
25000 We will continue work on color region analysis which is based on
25100 proximity rather than contiguity. Another effort combines region
25200 analysis with a decision theoretic approach based on a world model.
25300 We anticipate that these modules will aid the integrated system to
25400 function in more complex environments than have been attempted in the
25500 past. There is a growing effort on navigation of vehicles in outdoor
25600 scenes. While this is not directly relevant to the industrial
25700 environment, it does reflect the growing involvement in complex
25800 visual tasks, and will lead to further development of an integrated
25900 visual system.
26000
26100 C. MANIPULATOR DEVELOPMENT
26200
26300 Manipulator research began at Stanford Artificial Intelligence
26400 Laboratory in 1965 with the acquisition and programming of a
26500 prototype Rancho Orthotic Arm. A later version of this arm was used
26600 in vision directed block stacking programs.
26700
26800 Our present arm, the result of a continuing design project, replaced
26900 the Rancho Arm in 1970. This arm like its predecessor was controlled
27000 by a digital servo program running under the time sharing computer.
27100 The new arm was used in the "Instant Insanity" demonstration in 1971.
27200 A second arm of the same type, with only minor modifications, is
27300 currently being interfaced to the computer. Both arms will share a
27400 common work space and will be able to be used individually, or
27500 together for two handed tasks.
27600
27700 Each joint of the arm is calibrated in terms of torque versus drive
27800 level and it is possible to predict the joint drive levels to produce
27900 a given force at the hand. Further, for the hand to exhibit degrees
28000 of freedom of motion, various joints are selected to be "free", that
28100 is to have no position feedback applied to them. By exerting a force
28200 at the hand and providing that certain degrees of freedom exist it is
28300 possible to perform a much wider class of tasks than simple
28400 positional control, for instance screwing in screws, mating parts,
28500 inserting tools and turning cranks.
28600
28700 Touch sensors on the fingers allow the hand to sense objects without
28800 disturbing them. When activated the arm will stop when any of the
28900 sensors touch. Branching arm instructions can test the state of the
29000 sensors and modify the execution of an arm program. When screwing a
29100 nut onto a bolt, for instance, the hand is first centered over the
29200 bolt by touching it.
29300
29400 Control of the arms is by a series of high level instructions; these
29500 instructions may be executed directly, or may be assembled together
29600 into an arm program. Arm programs may then be executed, as desired,
29700 by the control program. The control instructions may be accessed
29800 directly by other programs, using the message procedure mechanism, or
29900 in the form of symbolic source file.
30000
30100 Positional accuracy at the hand is typically 0.1 inches with
30200 repeatability of 0.02 inches. Force control at the hand is within 10
30300 ounces. The data rate to run the arm from an assembled arm program is
30400 of the order of 100 bits/second.
30500
30600 Manipulator research moves ahead simultaneously on three fronts:
30700 control, primitives and devices. Tasks are translated into an
30800 existing hand language and where the task cannot be accomplished in
30900 terms of existing primitives, new primitives are formulated in a
31000 consistent manner. When a new primative is formulated, real time
31100 routines are developed to perform the action. This also leads, in
31200 many cases, to the development of new hardware. Also as new
31300 developments occur in hardware design, existing real time routines
31400 are either re-written or modified.
31500
31600 Research is now concentrating on the following areas:
31700
31800 Strain gauge instrumentation of the wrist, in order to allow for more
31900 sensitive interaction of the hand with a task. High resolution,
32000 high sensitivity touch sensors, to increase tactile recognition and
32100 control abilities. These developments will lead in turn to new
32200 control concepts.
32300
32400 Development of tools for the hand. These tools can be instrumented
32500 such that the computer can feel and or see at the end of the tool.
32600 This should be a very important area of research as it goes beyond
32700 present human capabilities.
32800
32900 Two handed arm control, the planning and execution of tasks requiring
33000 the coordinated motion of both arms. This will require an extension
33100 of the existing servo program and the analysis of synchronization
33200 problems. The development of collision avoiders, so that arms will
33300 not collide with other objects and with each other.
33400
33500 Modification of the existing arm design to provide for better
33600 incremental motions, more degrees of freedom and a more rugged design
33700 suitable for industrial use. Development of arm and eye computer
33800 interfaces. These contain synchronized analog/digital converters
33900 for input and current controlled drivers for output.
34000
34100 The development of new hands is expected as more complex tasks are
34200 attempted. At present, with only one hand, we have been performing
34300 the positioning type of hand actions. When the second arm is
34400 operational we will be able to perform holding and pulling actions as
34500 well.
34600
34700 D. SYSTEM REQUIREMENTS FOR ADVANCED AUTOMATION
34800
34900 A key element in the development of any large programming effort is
35000 the software support. There has been continual effort along these
35100 lines at the Stanford Artificial Intelligence Laboratory and many
35200 useful ideas have been developed. We propose to extend the research
35300 along two separate lines: support for research and system software
35400 for operating automated systems. We will briefly describe our past
35500 efforts in this area and what we propose to undertake.
35600
35700 The Stanford Artificial Intelligence Laboratory has pioneered in
35800 several areas of system programming, including timesharing systems,
35900 interactive graphics, and programming languages.
36000
36100 A great deal of additional effort has gone into direct support of
36200 hand-eye research. Very early, we established a set of standard
36300 formats and library routines [Pingle 72a]. A programming language,
36400 SAIL, was implemented in 1968. The language incorporates special
36500 features for large real-world-oriented problem solving systems and is
36600 currently in use at a number of Artificial Intelligence and other
36700 laboratories.
36800
36900 As we began to assemble larger and more sophisticated collections of
37000 programs, we felt a need for flexible mechanisms for cooperating
37100 sequential processes. The resulting monitor and language features
37200 [Feldman & Sproull 1971] have worked well for us and have been
37300 influential on other projects. Recently, we have added an
37400 additional set of features to SAIL to facilitate non-deterministic
37500 and event-directed programming. The early reactions to this work,
37600 both within the laboratory and without has been very encouraging.
37700
37800 We expect relatively little work to be needed on further systems
37900 support for basic research. There will undoubtedly be efforts to
38000 take advantage of new hardware and of new programming ideas that may
38100 develop in the course of this program. Almost all of this work
38200 should be covered by other funding.
38300
38400 The development of software for the assembly station will require
38500 more of our effort. The central-local computer configuration
38600 requires code to be resident in the local computer and in the shared
38700 machine. The low level programs in the local machine will not be
38800 modified often, thus we can assume that a compiler resident in the
38900 central machine will suffice. The planning programs and general
39000 vision programs will be resident in the shared system. There are two
39100 possible configurations for the shared computer system: A PDP-10
39200 equivalent machine or a software package in ASA Standard Fortran IV
39300 which will run on any sufficiently large machine. One of the tasks
39400 under this proposal would be to evaluate these two alternatives.
39500
39600 A major share of the programming research effort will be concentrated
39700 on developing a programming system for automated assembly. This will
39800 combine ideas from our past work on languages, representation,
39900 computer graphics and arm control. We view this task as a branch of
40000 the larger effort on automatic programming being undertaken in this
40100 laboratory and elsewhere.
40200
00100 III. UTILIZATION PLAN-DISSEMINATION OF RESULTS AND RELATION WITH INDUSTRY
00200
00300 All of the discoveries made under this grant will, of course, be made
00400 public. We view our role in information transfer in a much broader
00500 way. Having no desire to sell products and services and no clients
00600 with interests to protect, a university laboratory should be
00700 primarily a place for acquiring and spreading knowledge.
00800
00900 The Stanford Artificial Intelligence Laboratory has always considered
01000 information transfer to be one of its primary goals. In addition, to
01100 our published papers and reports, we have made a number of film
01200 reports which have been extensively circulated. There have been
01300 thousands of visitors, ranging from elementary school classes to
01400 major scientific, governmental and industrial missions from many
01500 countries. We have also encouraged extended visits by workers from
01600 other laboratories and from industry. Members of the laboratory staff
01700 have played a major role in the organization of journals, conferences
01800 and seminars in artificial intelligence. Many of our algorithms,
01900 systems, and hardware designs have been adopted by other groups.
02000
02100 The success of the program proposed here depends heavily on
02200 interaction with industry. We will need additional information on
02300 the opportunities for automation and will try to involve manfacturers
02400 in the development of the pilot assembly station. We have good
02500 working relationships with a number of firms and we expect that a
02600 concrete project will help forge the bonds necessary for the transfer
02700 of advanced automation technology from the laboratory.
02800
02900 One promising approach to technology transfer is a staff associates
03000 program. This involves residence of about a year's duration at the
03100 laboratory by industrial scientists. Such a program was used
03200 successfully at the MIT, Electronic's Research Laboratory in the area
03300 of programming systems for numerically controlled tools and design
03400 automation. We intend to utilize this as well as other mechanisms for
03500 gaining the insights that industrial experience can add to our
03600 enterprise.
03700